There is an inherent problem with design, which - to me - also coincides with the key differences between what I know of about US and Russian military design during the Cold War. See, the US is an incredibly wealthy nation with a lot of money lying around - no political commentary here, please - and so when the folks designing U.S. military technology ran into a problem the solution was, generally, to just throw more money at it. As such, you ended up with very high performance, cutting edge planes and vehicles that also needed a lot of maintenance and money to keep running. The Soviets on the other hand, didn't have as much money, and so they had to rely on ingenuity and down right craftiness. They had to compete with the high level gear the U.S. was coming out with, but also keep costs down. The end result were planes that were a bit lower in performance spec (at times, and in places, and obviously ymmv), but didn't need as much work to keep going.
How does this apply to design? Well, once you've started to make your own systems, it becomes very easy to go "I want to run game X, but what system? I know, I'll just make one." You stop looking for the system that can do your game right, and instead want to make one. The problem with that is, making a game system is a lot of work. Sure, if you put the work into it, you can have a working alpha/prototype for a game in a couple of days (I know, I've done it.) However, that alpha is going to have issues and is going to need time to have things hammered out and put into place. All the while, one of those books on your shelf may have the perfect answer for you.
Why You Shouldn't
Now, I'm not saying that new games are bad. In fact, we'll get to the good parts of it in a second, but there is also consideration that you need to go through before you roll up the sleeves and get to work. First and foremost, are you just re-inventing the wheel? Secondly, is there already a tool out there that does the job? A lot of smart design is using the rules that you already have, rather than trying to work up something new.
For example, the GM in the Dark Heresy game I am in, didn't like how Bolters worked. They didn't do their job (killing Space Marines) as well as they should. However, rather than add damage to the weapon, he looked through the rules and gave all Bolters the "Felling" trait, which makes it better against things like Space Marines, but not making it a god weapon against non-Space Marines. In Deathwatch, the designers solved the same problem by adding a damage die to the Bolter's damage, instead of using Felling. As such, Bolters are perhaps more - depending on your view - than they should be against some targets. Neither of these is the 'wrong' answer, but in one the person used rules that were there, and in the other they added something to make up for it. Obviously, making this point, I think one is the more elegant (if more complex) choice.
The same is true for making a whole new system. Wouldn't you be better of just looking through Hackmaster or D&D before making a new fantasy based, tactical combat dungeon crawler game? Or looking at BattleTech before making a tactical mecha game? Not that you can't still build, but the tools you need may already be there.
Why You SHOULD
Enough of the negativity, there are also reasons you should. Now, I said in the beginning that all these genres and settings have already been covered, but there are still nuances to be found in them and other points of view that may need to be worked on. I have two samurai RPGs sitting on my shelf, designed by the same guy even, and both of them push forward different concepts and view points for handling things. Depending on the kind of Samurai game I wished to run, I could use one or the other, and each has their own merits.
More to the point,t here are specific elements that you can bring to the table with your own design. Problems you see in other games you can address, and new solutions to old problems you can put forward. This is how the industry grows, and how the medium changes. Where would we be if the FATE system hadn't been made because the "Storytelling System" already existed and worked fine?
People sometimes say that you're re-inventing the wheel when you start this process. In other words, you're doing something that doesn't need to be done, as the wheel already exists and all that stuff. However, take a moment and look out on your nearest street. How many different kinds of wheels do you see on those cars? Now compare those to wagon wheels, and bicycle wheels, and wagon wheels. There's lots of room for different wheels.
In Short: Re-invent the wheel all you want, just be doing it for the right reasons.